Anthropic Revolutionizes Computing with Its Autonomous AI Agent Claude

Anthropic Unveils a New Era of Computer Control with Claude

On March 23, Anthropic took a giant leap in the field of artificial intelligence by launching the Dispatch mode of its agent Claude. This new tool allows for autonomous and remote control of a computer directly from a phone. This advancement follows the successful integration of Copilot Cowork via a mobile application and marks a significant milestone in the automation of computing tasks.

How Dispatch Mode Works

To activate Dispatch mode, you need to synchronize your Claude account on two distinct devices. The procedure is simple and quick:

• First, open the Claude application on your desktop.
• Next, log in with the same account on the Claude mobile application.

Once this connection is established, your phone becomes a powerful remote control, capable of sending instructions to your computer, even from a distance. The user can send commands from their phone, which Claude will execute on the desktop. The agent prioritizes direct connections with services such as Slack or Google Calendar. In the absence of direct integrations, Claude takes control of the screen, using the mouse, keyboard, and browser to accomplish its tasks. It can open files, browse the internet, and launch development tools autonomously. However, Claude requires explicit permission before accessing a new application.

In terms of availability, Dispatch is accessible on iOS and Android platforms via the Claude mobile application. On PCs running Windows, the agent is limited to Claude Cowork without full computer control, although it can interact with Chrome. Full computer control remains exclusive to macOS systems for now. To benefit from these features, it is necessary to subscribe to a Claude Max account (costing between $100 and $200 per month) or Pro ($20 per month).

Simplicity and Limitations of Tasks

Currently, Dispatch and computer control are primarily used for simple everyday tasks. Here are some examples of what Claude can accomplish:

• Add a PDF file to a Google Calendar invitation.
• Organize a folder of photos by date or theme by sorting them into subfolders.
• Create a PowerPoint presentation from a text document.
• Populate a spreadsheet with data from multiple files.

Security and Restrictions in Place

Anthropic presents this feature as a research preview, emphasizing that computer control is still slower than direct integration, and complex tasks may require multiple attempts. Additionally, the computer must remain powered on with the desktop application active for Dispatch to function properly.

Regarding security, Anthropic has implemented several protective measures. The system automatically analyzes model activations to detect any prompt injection attempts, and the user has the option to interrupt Claude at any time. Certain applications are also blocked by default to minimize risks, although Anthropic has not specified which ones. The company recommends not allowing the agent to handle sensitive data and starting with trusted applications.

Since the beginning of the year, Anthropic has been launching products at a rapid pace: context at 1 million tokens, interactive visualizations, code review in Claude Code, Dispatch, and now computer control. This strategy of releasing products early, even if imperfect, and gradually improving them allows the company to move closer to its vision of a truly autonomous personal assistant. A vision that is gradually materializing with a reliable and secure OpenClaw.